1. Field of the Invention
The present invention relates to a high-speed communication system and, more particularly, to a high-speed communication system with a feedback synchronization loop.
2. Description of the Related Art
A gigabit communication system is a system that transfers billions of bits of data per second between the nodes of the system. Gigabit communication systems commonly handle the data transferred over the backplane of the internet, and are expected to handle the data transferred between next-generation processors and peripherals, such as hard drives and printers.
FIG. 1 shows a block diagram that illustrates a conventional gigabit communication system 100. As shown in FIG. 1, system 100 includes a high-speed transmission medium 108, such as a fiber optic cable, and a number of communication devices 110 that receives data from, and transmits data to, medium 108.
Each communication device 110, in turn, includes a physical layer device 112 that is connected to medium 108, and a processing device 114 that is connected to physical layer device 112 by a number of lines 116. Physical layer device 112 includes a serializer/deserializer (serdes) that transforms data received from medium 108 into a signal format that is compatible with processing device 114, and transforms data from processing device 114 into a signal format that is compatible with medium 108.
When transferring data to, and receiving data from, processing circuit 114, the serdes typically utilizes a data signal which has a logic high that is represented by a maximum voltage which is equal to the supply voltage used by the processing circuit.
For example, when device 114 is formed in a 0.35 micron photolithographic process, physical layer device 112 transmits data to, and receives data from, device 114 with data signals that have a maximum voltage of approximately 3.3V, the supply voltage commonly used with 0.35 micron devices.
One channel of data is typically transported across medium 108, and between physical layer device 112 and processing device 114, at 1.25 Gb/s, with speeds of 2.5 Gb/s under consideration.
Processing device 114 processes the data received from medium 108 by physical layer device 112, and outputs processed data to physical layer device 112 for transmission onto medium 108.
Physical layer device 112 and processing device 114 are typically encapsulated in separate chips which are placed on the same printed circuit board due to the largely analog nature of device 112 and the largely digital nature of device 114. One consequence of this approach, however, is that electromagnetic interference (EMI) requirements limit the maximum speed that data can be exchanged between devices 112 and 114.
For example, when data is exchanged between devices 112 and 114 with data signals having a maximum voltage of approximately 3.3V, the maximum speed that can be obtained without exceeding the EMI requirements is approximately 125 Mb/s.
Thus, to handle one channel of inbound data, which is received at 1.25 Gb/s, 10 inbound lines 116 are required to transport data from device 112 to device 114, where physical layer device 112 has 10 output ports and processing device 114 has 10 input ports. (10 inbound lines 116 at 125 Mb/s provide one channel of inbound data at 1.25 Gb/s).
Similarly, processing device 114 requires 10 outbound lines 116 to transport one channel of outbound data from device 114 to device 112, where processing device 114 has 10 output ports and physical layer device 112 has 10 input ports. Thus, device 112 and 114 each require 20 input/output ports, with 20 corresponding pins, to handle the inbound and outbound data for one channel.
To provide additional EMI margin and greater chip-to-chip spacing, communication devices with reduced chip-to-chip speeds are also available. These reduced-speed devices typically transfer data between devices 112 and 114 at 62.5 Mb/s.
One problem with communication devices that have reduced chip-to-chip speeds, however, is that devices 112 and 114 have twice as many I/O ports and twice as many pins. Thus, with a reduced-speed device, devices 112 and 114 require 40 pins each (20 inbound lines 116 at 62.5 Mb/s are required to provide one input channel at 1.25 Gb/s, while 20 outbound lines 116 at 62.5 Mb/s are required to provide one outbound channel at 1.25 Gb/s).
The pin problem becomes even worse when devices 112 and 114 are packaged as four and eight-channel devices. When packaged in this way, devices 112 and 114, when operating at a high chip-to-chip speed, i.e., 125 Mb/s, each require 80 pins and 160 pins to support four and eight-channel devices, respectively. Further, devices 112 and 114, when operating at a slower chip-to-chip speed, i.e., 62.5 Mb/s, each require 160 pins and 320 pins to support four and eight-channel devices, respectively.
The pin problem reaches critical stages when devices 112 and 114 are scaled up to handle a 2.5 Gb/s data rate from the current 1.25 Gb/s data rate. At these higher speeds, devices 112 and 114, when operating at a high chip-to-chip speed, i.e., 125 Mb/s, require 160 pins and 320 pins to support four and eight-channel devices, respectively. Further, devices 112 and 114, when operating at a slower chip-to-chip speed, i.e., 62.5 Mb/s, require 320 pins and 640 pins to support four and eight-channel devices, respectively.
Thus, there is a great need to reduce the pin counts of devices 112 and 114 when devices 112 and 114 are scaled up to handle a 2.5 Gb/s data rate. (In addition to consuming huge amounts of silicon real estate, large pin count devices also consume large amounts of power.)
One conceptual approach to reducing the pin counts is to exchange data between devices 112 and 114 with a single-ended signal that has a lower maximum voltage. For example, by lowering the maximum voltage of a single-ended data signal from 3.3V to 500 mV, the frequency of the data signal can be increased from 125 Mb/s to approximately 1.25 Gb/s without exceeding the EMI requirements. By lowering the maximum voltage from 3.3V to 250 mV, the frequency of the data signal can be increased from 125 Mb/s to approximately 2.5 Gb/s without exceeding the EMI requirements.
One problem with this conceptual approach, however, is that it is extremely difficult, if not impossible, to form inbound detectors on processing device 114, and outbound detectors on device 112, that accurately detect logic ones and logic zeros from a single-ended gigahertz data signal that has a maximum voltage in the hundreds of millivolts due to the voltage margins required by the detectors.
Another problem with this conceptual approach is that much more complex clock recovery circuitry is required to recover a clock signal from a data signal operating in the gigahertz range, such as 2.5 GHz, than from a data signal operating in the megahertz range, such as 125 MHz. Thus, much of the clock recovery circuitry that is utilized in the serdes would also be required in processing device 114 to recover the clock from a gigahertz data signal (output by device 112 to device 114) that has a maximum voltage in the hundreds of millivolts.
Another approach to reducing the pin count, that also avoids this duplication, is to integrate the functions of physical layer device 112 and processing device 114 on a single chip. One problem with this approach, however, is the incompatibility of high-precision analog circuits, which make up most of the circuits on physical layer device 112, with digital circuits, which make up most of the circuits on processing device 114.
One of these incompatibilities is the speed with which new processing technologies can be implemented. For the present, digital circuits are easily adapted to new (and smaller) processing technologies because the voltage levels that represent logic ones in the new processing technologies are still easily distinguished from the voltage levels that represent logic zeros.
For example, in both a 0.5 micron photolithographic process and a 0.35 micron photolithographic process, where a logic one is represented by a 5V signal and a 3.3V signal, respectively, the logic one is easily distinguished from a logic zero which, in both cases, has a voltage near zero.
For high-precision analog circuits, however, moving from a 0.5 micron photolithographic process to a 0.35 micron photolithographic process, where the supply voltage drops from 5V to 3.3V, dramatically reduces, among other things, the dynamic ranges of the analog devices. Further design (and time) is then often needed to develop devices which operate in these ranges.
Thus, integrating the functions of physical layer device 112 and processing device 114 on a single chip increases the time required for the digital circuitry in the integrated device to take advantage of the reduced size and power requirements provided by a new photolithographic process.
As a result, there is a need for a communication device that has a physical layer device and a processing device which operate in the gigahertz frequency range with substantially fewer pins.